April 2026

Citizens, Infrastructure, and Knowledge in South Africa's Draft AI Policy

The Verdict
Hybrid, Leaning Derivative

The policy is a selective adaptation layered onto a substantially imported architecture.

South Africa's AI policy trajectory represents a hybrid effort, but one whose centre of gravity sits closer to the derivative end of the spectrum than to genuine origination. The 2026 Draft Policy makes real and sometimes inventive attempts at contextual grounding: anchoring AI governance in the Constitution, invoking Ubuntu, and foregrounding indigenous language preservation.

However, the core policy architecture (risk-based classification, ethical principles, and innovation-competitiveness framing) remains fundamentally imported.

The result is a document that reads global and annotates local, rather than one that thinks from local conditions outward.

Analytical Framework

A Blueprint for Localization

While the current SA Draft Policy relies heavily on imported, top-heavy institutional blueprints, a genuinely contextual AI strategy requires ground-level execution.

The following seven pillars outline my recommended framework for how South Africa must translate global ideals into actionable policies across National, Provincial, and Local tiers.

01. The Citizen

AI policy is ultimately a question of whose life it touches. The citizen is not a passive data subject but a current or possible innovator and rights-bearer entitled to explanation, redress, privacy, and meaningful participation under the Constitution.

National
Legislate algorithmic due-process rights and statutory redress pathways.
Provincial
Ensure service-delivery AI systems comply with Batho Pele principles.
Local
Run community consultations in multiple official languages before implementation.

02. The Infrastructure

No AI policy survives contact with an unstable grid, thin fibre coverage, or foreign-owned compute. Infrastructure is a hard constraint.

National
Coordinate energy, spectrum, and compute strategy; negotiate sovereign-cloud terms.
Provincial
Align broadband and data-centre incentives with the national plan.
Local
Municipal fibre, public Wi-Fi, and last-mile connectivity to close access divides.

03. State Capacity

A risk-classification regime borrowed from Brussels presumes a regulator that can audit, inspect, and enforce. A deep institutional capacity check is required.

National
Fund technical AI-auditing capacity inside regulators; avoid creating unfunded bodies.
Provincial
Build AI literacy in provincial treasuries and procurement units.
Local
Train municipal officials to evaluate vendor claims and refuse opaque tech.

04. The Economy

Over a third of South Africa's workforce operates outside the formal economy. An AI policy that speaks only to large firms writes millions out.

National
Labour-market transition policy, platform-worker protections, and SMME programmes.
Provincial
Sector-specific reskilling tied to provincial economic profiles.
Local
Support township digital hubs and informal-trader digitisation.

05. Knowledge & Language

Twelve official languages and indigenous knowledge systems determine whether AI is built for South Africans or merely deployed on them.

National
Fund indigenous-language corpora, open datasets, and doctoral pipelines.
Provincial
Integrate AI literacy into basic and higher-education curricula.
Local
Public libraries and community centres as major access points for tools.

06. Governance & Accountability

Without a clear inter-governmental architecture, the framework risks becoming a national document that never reaches the municipal coalface.

National
Clarify lead department, inter-ministerial coordination, and continental alignment.
Provincial
Provincial AI focal points accountable to legislatures.
Local
Mandatory AI-use registers for municipalities, open to public inspection.

07. Innovation, Citizens & Context

Innovation is treated as an outcome if ethics are addressed, missing its role in political-economic design and local knowledge structures.

National
Fund citizen-led R&D, public-interest model development, and fair-use copyright.
Provincial
Procurement preferences for locally produced models and datasets.
Local
Township innovation hubs with real stakes.

08. Build & Transform

The Framework nowhere commits the state to building a sovereign foundation model or to reorganising the machinery of government around AI-enabled decision-making. It names every argument for doing so (data sovereignty, foreign-provider dependence, indigenous-language under-representation), then declines to prescribe the treatment. The document imagines AI as something the state will oversee, not something the state will become. This is the difference between being a consumer and a producer of AI, and the Framework, for all its ambition, sits squarely on the consumer side.

National
Commit public funding to a sovereign South African foundation model built for local languages and contexts; create a central digital service with real authority over line departments rather than only advisory bodies.
Provincial
Rewrite procurement law so AI commitments cannot be captured by a handful of foreign vendors; embed data stewards inside provincial departments rather than outside them.
Local
Pilot genuine institutional redesign in at least one metro. Not a chatbot bolted onto a website, but a rethinking of how decisions get made, audited, and contested.

Comparative Findings

Key structural similarities and differences across the international texts and the SA frameworks.

Convergence of Terminology: Terms such as "human-centric," "trustworthy AI," "transparency," "accountability," "fairness," and "risk-based approach" appear across every text. This reflects a unified language template originating primarily from the OECD AI Principles and UNESCO.

Risk-based classification forms the structural backbone for the EU AI Act, the UN Model Policy, and the SA Draft Policy. Each relies on regulating through harm grading.

Governance architectures display mirrored elements: dedicated oversight bodies, ethics boards, and audit obligations are pervasive globally.

Legal Bindingness & Jurisdiction

The EU AI Act is sweeping binding legislation. The SA Draft Policy is a framework that may lead to legislation down the line. The UN documents remain purely guidance mechanisms.

Developmental Framing

Only the SA documents repeatedly frame AI primarily as a development tool destined to address poverty, unemployment, and inequality.

Cultural Anchoring

The SA documents explicitly invoke Ubuntu, setting them conceptually apart from the purely individual rights-based structure of the EU.

SA Draft National AI Policy (March 2026)
An 86-page policy document proposing extensive institutional bodies (Commission, Ethics Board, Authority, etc.). It cites the EU AI Act and GDPR "as a model," containing pre-regulatory open "OPTIONS."
SA National AI Policy Framework (October 2024)
A slim 14-page predecessor. It uses a Futures Triangle model but operates predominantly at the slogan level, lifting strategic pillars heavily from global frameworks.
EU AI Act (March 2024)
The world's foremost comprehensive and binding AI legislation. It focuses on internal market integration and rigid systemic evaluation.

AU Strategy: Convergences

Both SA and the AU frameworks share a definitive developmental orientation, emphasizing the "AI divide" and harnessing technical benefits amidst historical constraints.

AU Strategy: Divergences

The AU Strategy is aggressive about African-centric AI capability and cultural renaissance. SA acknowledges these but predominantly leans on OECD regulatory vernacular rather than fully embracing continental disruption.

Discourse & Interpretation

Evaluating the authoritative levers and core narrative mechanisms driving the policy documents.

Legitimizing References
The SA Draft Policy asserts its legitimacy through a dual strategy: referencing internal constitutional anchoring (Bill of Rights, POPIA) and international benchmarking (EU AI Act, OECD). Consequently, the text embodies a unique hybrid posture.
"The guidelines espoused in this policy must be followed. And in line with the OECD's human-centred values..." SA Draft Policy, Section 4
Implied Audiences & Power Assumptions
Citizens function in only three capacities in the text: beneficiaries of services, victims to be protected, or consumers of awareness campaigns. The active policy machinery is structured from the top down, overlooking foundational capacities for localized, citizen-led governance models.
Hermeneutic Reality
Fundamentally, the SA policy is a manifestation of state aspiration burdened by resource limitations. It advocates for global-scale institutional infrastructure from within a state that faces vast existing administrative and physical realities. Even well-intentioned elements like Ubuntu serve mostly as linguistic accents over a Western European regulatory template. Its brightest conceptual spark exists in highly specific ideas like the AI Insurance Superfund.

Borrowed vs. Grounded Assessment

Diagnostic breakdown of specific thematic modules within the policy and their degree of local grounding.

Filter
Theme Classification Analytic Take
Risk-Based Regulation Imported Overtly extracts the EU AI Act's structural logic globally transposed.
Ethical Principles Imported Nearly verbatim mimicry of prevailing OECD and major corporate principles.
Inclusion & Equity Adapted Integrates SA's legacy of inequality into global inclusiveness dialogue.
Constitutional Anchoring Grounded Specifically intertwined with provisions from the SA Bill of Rights and domestic tech law.
AI Insurance Superfund Grounded Adapted off the framework of the localized Road Accident Fund concept.
Ubuntu Philosophy Performative Invoked purely on an aspirational values basis without structural mechanics.
Education System Performative Ignores baseline disparities in the current school system while projecting AI ubiquity.
Digital Infrastructure Ambiguous Recommends vast technological networking completely detached from persistent grid instability.
Informal Economy Unsaid Total omission of an economic engine involving millions of marginalized participants.
Fiscal Feasibility Unsaid Broad mandates built on absent budgeting metrics.

Critical Silences

A framework defined as much by its exclusions as its written ambitions.

State Capacity & Implementation Reality

The policy mandates creating seven brand-new institutional regulatory bodies-all while existent regulators face overwhelming operational constraint. Institutional blueprinting frequently overshadows execution capabilities.

Grid Infrastructure

Data centres, digital sovereignty, and complex computing architectures run sharply counter to South Africa's deep realities regarding load shedding and electrical grid decay. This critical constraint is profoundly understated.

The Informal Labour Base

A staggering percentage of SA's workforce relies on the informal economy. And yet, this sector has no seat at the table concerning automation vectors, economic shockwaves, or transitional protections driven by AI propagation.

The Absent Sovereign Model

Nowhere in the Framework does the state commit to building a national foundation model. It worries about data sovereignty, foreign providers, and the under-representation of South African languages, and then proposes no domestic model to address any of it. The UAE built Falcon, France backed Mistral, Singapore funded SEA-LION, India is funding Indic LLMs; South Africa regulates and enables but does not build. The document names every disease the treatment would cure, and then declines to prescribe it.

Institutional Transformation of the State

The Framework proposes many new bodies (an Ethics Board, a National AI Commission, a Regulatory Authority, a Safety Institute, an Ombudsperson), yet all of them sit above the machinery of government, watching it. None are inside the departments rebuilding how decisions are actually made. Serious AI-driven governance implies central digital services with real authority, data stewards embedded in ministries, rewritten procurement law, and new civil-service training. The document treats AI as a tool to procure rather than an occasion to reconstruct public capacity.

Final Synthesis

A Derivative Translation

South Africa is actively engaging in policy creation, translating modern global AI governance discourse into its local institutional syntax. This constitutes a valuable but undeniably derivative enterprise.

The profound and particular realities of the nation-informal economics, profound energy deficits, structural limits in fiscal policy, and immense bureaucratic stresses-must generate the shape of policy formulation. Allowing core values like Ubuntu to manifest as concrete operational tools will begin defining a grounded, rather than merely annotated, national path forward.

Works Cited

1. Republic of South Africa, Department of Communications and Digital Technologies. National Artificial Intelligence Policy Framework. Government Gazette No. 54477, 10 April 2026. Pretoria: Government Printing Works.

2. On the economics of scraping and journalism, see Reuters Institute for the Study of Journalism, Digital News Report (annual), and contemporary debate on text-and-data mining exceptions in copyright law.

3. African Union Commission. Continental Artificial Intelligence Strategy: Harnessing AI for Africa's Development and Prosperity. Addis Ababa, July 2024.

4. Pritchett, L., Woolcock, M., & Andrews, M. (2013). Looking Like a State: Techniques of Persistent Failure in State Capability for Implementation. Journal of Development Studies, 49(1), 1–18.

5. Andrews, M., Pritchett, L., & Woolcock, M. (2017). Building State Capability: Evidence, Analysis, Action. Oxford: Oxford University Press.

6. Supplementary: UNESCO, Recommendation on the Ethics of Artificial Intelligence (2021); OECD AI Principles (2019, updated 2024); European Parliament, Regulation (EU) 2024/1689 (AI Act).

Share